Yet another issue -- or perhaps it is tacit in
formulations of other issues -- is that it is
easy to find yourself making an assumption that
different states are mutually independent. It
could be that all influences on this syndrome
(right word?) arise within-state, but if
geography has any meaning then at least some influences
(demographic, social, economic, ...) will show
spatial autocorrelation. That is,
Michigan is more like Wisconsin than say Alabama.
This tends to mess up P-values _and_ confidence intervals.
Nick
[email protected]
Brent Fulton
> Thank you all for your feedback.
>
> I have individual-level data from the US on whether a child has been
> diagnosed with ADHD (attention deficit/hyperactivity
> disorder). The data is
> from a complex survey design where each child has a
> probability of selection
> weight and each stratum is a state, so state-level estimates
> are valid.
>
> My research question--better stated from Steve Samuel's suggestion--:
> First null hypothesis structure:
> Ho: proportion of US's children with ADHD minus proportion of
> state's (e.g.,
> Michigan's) children with ADHD = 0
> Using my sample, I wanted to use a statistical test to
> determine whether or
> not to reject the null hypothesis, where there would be 50
> null hypotheses
> (one for each state). Michael Frone's Tuesday, December 05,
> 2006 8:27 AM
> (PST) email provides a method for this. (Need to further
> investigate Austin
> Nichols's reply.)
>
> The second null hypothesis structure is:
> Ho: proportion of e.g., non-Michigan's children with ADHD
> minus proportion
> of e.g., Michigan's children with ADHD = 0; As above, there
> would be 50 null
> hypotheses (one for each state).
> There are many methods to test this null hypothesis in Stata. From a
> practical standpoint, the decision whether to reject the null
> hypothesis
> using the first null hypothesis structure for a particular
> state will likely
> be the same as the decision for the second null hypothesis structure.
>
> I then plan to use an indirect-adjustment method to adjust
> the states' ADHD
> prevalence rates for different child-level characteristics
> across states,
> and re-run the hypotheses tests with the adjusted rates.
>
> Your input has provided me methods for the first null hypothesis
> structure--which I didn't have before. And since the scope of this
> issue--which null hypothesis structure is typically used in
> the literature
> and the strengths/weaknesses of each--is getting beyond Stata
> usage, I don't
> want to burden this email list. After thinking about it more,
> I think the
> second null hypothesis structure is cleaner since I can come up with
> examples that would make the first structure answer a very
> odd question.
> Maybe the literature (that I have read) that states they are
> using the first
> structure are, in fact, actually using the second? But if you have
> suggestions for me personally, I'd welcome them.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
